Skip to main content

Understand Session Management and Memory

This guide explains how the GlobalAI platform handles conversation history and how you can track agent interactions through the auditing system.

Memory & context retention​

By default, Large Language Models (LLMs) are stateless. This means they don't "remember" previous interactions. To maintain context over multiple messages, configure memory for the agent.

Stateless behavior (no memory)​

If you don't configure memory for an agent, the system treats every message as a new, isolated session.
The agent only considers the current prompt and has no knowledge of previous messages or answers.

Stateful behavior (with memory)​

When you attach memory, the agent keeps the context of the conversation.
It can recall:

  • Previous user messages
  • Its own prior responses

This lets the agent support follow-up questions and multi-turn conversations.

Shared memory in Workflows​

In complex workflows, you may have several agent nodes that need to share context.
For example, one agent may gather data, and a second agent may analyze it based on previously collected information.

The recommended solution is to use a shared SQLite database inside the workflow’s working directory.

How it works​

  • File-Based Storage: Configure the first agent to store memory in a temporary SQLite file (for example, db.sqlite) located in the working directory.
  • Passing Context: Attach the same SQLite database file to subsequent agents in the workflow so they can read the existing memory.
  • Session Linking: Use a static Session ID (such as the workflow's Execution ID) across agents to let them continue the same "conversation thread."

This creates a shared brain across several workflow steps.

Auditing and observability​

The Audit Sessions panel provides deep observability into how your agents perform and interact with users.

Auditing only works if the agent has memory attached. If an agent runs without memory, the system doesn't store sessions and no audit logs will be available.

Scope of auditing​

The audit log captures all interactions, whether they:

  • Occur inside a workflow execution, or
  • Come through the external API.

Key metrics available​

For each session, you can access:

  • Conversation History: Full transcript of messages exchanged between the user and the agent.
  • Tool Usage: How many times the agent called external tools during the interaction.
  • Performance Metrics: Token consumption (cost tracking) and total execution time.

Now that you know how session management and memory work in GlobalAI, you might want to learn more about:

Understand AI agents

Learn the core concepts behind AI agents, including their components and how they function within GlobalAI.

How to Use an AI agent in a Workflow

A step-by-step guide to building and deploying AI agents in GlobalAI.

Manage Knowledge Bases

Learn how to add and configure knowledge bases for your AI agents in GlobalAI.